https://sunlightrecovery.com/social-media-algorithm-mental-health/
Edited by Mason Gunawan, Madelyn Bello, Amelia Cantwell, & Owen Andrews
Social media algorithms are unpredictable and are developing faster than humans can keep up. In October 2023, 51 attorneys general sued Meta, claiming the company violated consumer protection laws through the use of addictive features on their platforms. Algorithms feed users content using “machine learning” software to decipher what content the user will enjoy and then recommend it. This can be developed through actions as simple as interacting with viral videos, and it keeps the users scrolling for as long as possible. Thus, algorithms are at the center of many social media concerns. Meta and the Attorneys General reached a settlement on August 26th, 2026, for more than $17 billion in damages and added restrictions to the app. These restrictions include bans on cosmetic filters for minors, an automatic school-hour and bedtime app lock, and a restriction on visibility of the number of “likes” or shares that one receives on a post. While this is a step in the right direction, the settlement also raises prominent legal questions, especially as social media algorithms continue to develop. Chief among them is whether algorithms should be treated as a product, and therefore be subject to product-liability law.
Product liability is the “legal responsibility imposed on any or all parties along the chain of manufacturing of any product for damage caused by that product.” Its purpose is to hold creators accountable for defective product design. Therefore, because social media algorithms create a foreseeable risk of harm to their users, they should fall under product liability scrutiny.
In June of 2020, the Third Circuit Court of the United States dismissed a complaint from a mother whose son was murdered by a man who had been granted pre-trial release based on New Jersey’s Public Safety Assessment (PSA). This assessment is used to aid in a decision about whether to detain someone pending trial by evaluating their “risk” level, or how likely they are to appear at future court dates and their risk or rearrest. The Public Safety Assessment, created by the Laura and John Arnold Foundation, used an algorithm to develop a risk assessment tool to help judges decide on pretrial release. In Rodgers v. Christie (2020), June Rodgers claimed that the PSA was a defective “product” and therefore should be subject to product-liability scrutiny. The Third Circuit Court defines a product as “tangible personal property distributed commercially for use or consumption”, dismissing Rodgers’ claim under the notion that the PSA was an “objective, standardized, and . . . empirical...risk assessment instrument” and not “tangible personal property” nor remotely “analogous to” it. Under the Third Circuit Court’s interpretation, the PSA does not qualify as a product under traditional product liability law because it was neither tangible or distributed commercially.
The Third Circuit Court’s decision in Rodgers v. Christie should not extend to all algorithms. Courts should consider algorithms as potential products under product liability law and examine whether defects in their designs contribute to foreseeable harm. Although algorithms are intangible, they can be developed and incorporated into products in ways that create foreseeable harm. Companies deliberatley design algorithms to perform specific functions, similarly to how a manufacturer designs a product. If the design of an algorithm creates a risk of foreseeable harm, its lack of tangibility does not eliminate the company’s responsibility for its design choices. Therefore, just because an algorithm isn’t a tangible item, it should not be automatically excluded from the definition of a product.
Although the Third Circuit Court excluded the PSA from product liability, the court’s reasoning should not mean that all algorithms are excluded from product liability. Social media algorithms actively choose what content the user gets pushed, whereas the PSA is used as a source of information for judges to consider when making decisions regarding pretrial releases. This distinction is important because social media companies exercise greater control over the content their users encounter, raising questions about who bears responsibility for harms created by the design of an algorithm.
If a company knows that its algorithm is capable of recognizing a user’s preferences and tailoring content to them accordingly, the decision to design an algorithm that repeatedly recommends young users to potentially harmful content becomes an issue of design rather than independent user behavior.
Unlike the PSA in Rodgers, which provides an algorithmic assessment for a judge to use for consideration in a pretrial release decision, social media algorithms have the power to directly shape what information the users receive. Social media algorithm design choices create risks, as presumptions are made about what content the user may enjoy based on previous interactions, without necessarily distinguishing between content that is enjoyable and content that may cause harm. As a result, algorithms may repeatedly expose users to harmful content, reinforcing insecurities or harmful thoughts and behaviors. For example, likes and follower counts trigger neural behavior that affects an adolescent's behavior and attitude. If a young adult is shown a post by a comparable person who has more likes than them, unnecessary tension and stress can be created because of sensitivity to social evaluation. Because algorithms are user dependent, they require information from the user’s own behavior, pulling into question whether harm was caused by the algorithm or the independent user. However, algorithms are constantly adapting to trends and updated user preferences. This makes it hard to apply product liability law to algorithms, as they do not have a relatively fixed design.
While these are important considerations, the ever-changing nature of algorithms should not automatically exempt them from product-liability law due to the risk of foreseeable harm. Liability should depend on whether the platform’s design contributed to the risk beyond what the user’s behavior alone would have produced. Users play a role in shaping the content they are pushed, but social media companies are in control of design choices surrounding what content is amplified. Rather than automatically excluding algorithms from the definition of product due to their intangible nature, courts should consider the function and design of algorithms before determining whether product liability law applies. This approach would not require companies to remove all potentially harmful content or hold them responsible for every user’s feed. Rather, it would allow courts to separate a distinction as to whether the company took reasonable measures to prevent such foreseeable harm created by algorithmic design. This would distinguish between harm resulting from independent user behavior and harm that a company’s design choices amplified. This approach would preserve the Third Circuit Court’s decision in Rodgers vs. Christie to exclude the PSA from product liability while also allowing product-liability law to address algorithms whose design contributes to foreseeable harm. As algorithms continue to evolve, courts must ensure that traditional definitions of a product do not become barriers to holding companies accountable for the harms their designs cause.